The Citation Audit

Every PubMed identifier on this site was checked against PubMed, one at a time. 21 of them were defective. Some pointed at a different paper entirely. Some pointed at a paper that had since been retracted, or at a study that administered a different drug from the one being graded. Some named a paper that, as far as EBL can establish, does not exist. All 21 are listed below, with what was cited, what that identifier actually resolves to, and what replaced it where anything did.

Every correction below is stored as a record in data.json, and this page is generated from those records at build time. It cannot show a correction that is not in the database, and it cannot hide one that is.

Read the percentage carefully, because the obvious way to compute it is wrong. The 21 corrections were all found between July 2026 and October 2026, in a re-audit of a corpus that had already been audited once. Dividing them by the 1858 identifiers the site renders TODAY would produce 1.1%, a figure that falls every time a citation is added and nothing is re-audited, which is why this page does not lead with it. The rate against the corpus actually audited is the figure on the left: 18 of the 489 identifiers this site carried in July 2026, measured by running the site's own reference counter over the archived snapshot for that date. A further 3 corrections were found after that date and are counted in the 21 above but not in that rate.

Measured, not remembered. refcount.populations()['rendered_unique'] from the current build, run over assets/data.json from the archived repository snapshot of July 2026, which already carries all 18 withdrawn_citations records including the retraction found that day. One definition of 'a PubMed identifier on this site', applied to two moments. An earlier version of this page put the figure for that date at 510; no count or combination of counts in that snapshot reproduces 510, so the measured 489 replaces it.

An earlier audit of the original citation set found a far higher error rate. That figure is not published here, because those citations no longer exist and it cannot be re-derived from this site. EBL does not publish numbers it cannot reproduce on demand, including its own.

The list above is one kind of defect: a citation taken off a record. The corrections below are the other kind, and they are the more common one. The citation stayed and what EBL said it showed was wrong. A paper filed as a null result that reported a significant effect. A trial and that trial's own subgroup counted as two independent studies. A grade resting on the replicated The same finding reproduced by a separate research group. A result seen once can be chance or bias; a replicated one is harder to explain away. route with nothing replicating it. None of these is visible to a check that asks whether an identifier resolves, because every one of them resolves. They are found by reading the paper against the sentence, and they are recorded here as they are found.

A PubMed identifier is a run of digits and carries no meaning. You cannot look at 30536195 and 30306467 and tell which is the paper being described. Neither can a reviewer, and neither can a reader, because checking costs a click and nobody clicks. The failure has a signature: a real identifier, a real journal, the right author named in the prose, and a paper about something else entirely. That is not a typo. A typo gives you a dead link.

Two of the corrections above are worse than wrong papers. They are attributions to papers that, as far as EBL can establish, do not exist.

EBL no longer checks citations by hand, because hand-checking is what produced this list. The build now refuses to publish the site if any graded item cites nothing, if any PubMed ID appears on a page without being recorded in the database, if any page argues for a grade it does not have, or if any claim the site makes about itself is not true of its own data. Those checks run on every deploy and they do not get tired.

Educational use only. Not medical advice. EBL grades published human evidence and does not recommend treatment. · Privacy · Disclosure · Terms · Accessibility

Evidence Based Longevity, Inc. is owner-operated and edited by Arnold Sandlow, D.C., a California-licensed chiropractor. Evidence Based Longevity publishes in English only · Machine translations are not produced or reviewed by EBL for accuracy

© Evidence Based Longevity, Inc.. All rights reserved